Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,78 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: cc-by-4.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-4.0
|
| 3 |
+
pipeline_tag: image-to-image
|
| 4 |
+
tags:
|
| 5 |
+
- pytorch
|
| 6 |
+
- super-resolution
|
| 7 |
+
- pretrain
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
[Link to Github Release](https://github.com/Phhofm/models/releases/tag/4xSPAN_pretrains)
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
[Neosr](https://github.com/muslll/neosr)'s latest update from yesterday included a [new adaptation of the multi-scale ssim loss](https://github.com/muslll/neosr/wiki/Losses#mssim_opt).
|
| 14 |
+
This was an experiment to test out the difference between making a SPAN pretrain with pixel loss with L1 criteria (as often used in research) vs mssim loss as its only loss.
|
| 15 |
+
Models are provided so they can be used for tests or also used as a pretrain for another SPAN model.
|
| 16 |
+
|
| 17 |
+
---
|
| 18 |
+
|
| 19 |
+
## 4xpix_span_pretrain
|
| 20 |
+
|
| 21 |
+
Scale: 4
|
| 22 |
+
Architecture: SPAN
|
| 23 |
+
|
| 24 |
+
Author: Philip Hofmann
|
| 25 |
+
License: CC-BY-4.0
|
| 26 |
+
Purpose: Pretrain
|
| 27 |
+
Subject: Realistic, Anime
|
| 28 |
+
Date: 10.04.2024
|
| 29 |
+
|
| 30 |
+
Dataset: [nomos_uni](https://github.com/muslll/neosr)
|
| 31 |
+
Dataset Size: 2989
|
| 32 |
+
OTF (on the fly augmentations): No
|
| 33 |
+
Pretrained Model: None
|
| 34 |
+
Iterations: 80'000
|
| 35 |
+
Batch Size: 12
|
| 36 |
+
GT Size: 128
|
| 37 |
+
|
| 38 |
+
Description: 4x SPAN pretrain trained on pixel loss with L1 criteria (as often used in research) on downsampled nomos_uni dataset using kim's [dataset destroyer](https://github.com/Kim2091/helpful-scripts/tree/main/Dataset%20Destroyer) with down_up,linear,cubic_mitchell,lanczos,gauss,box (while down_up used the same and with range = 0.15,1.5).
|
| 39 |
+
The new augmentations except CutBlur have also been used (since CutBlur is meant to be applied to real-world SR and may cause undesired effects if applied to bicubic-only).
|
| 40 |
+
Config and training log provided for more details.
|
| 41 |
+
|
| 42 |
+
---
|
| 43 |
+
|
| 44 |
+
## 4xmssim_span_pretrain
|
| 45 |
+
|
| 46 |
+
Scale: 4
|
| 47 |
+
Architecture: SPAN
|
| 48 |
+
|
| 49 |
+
Author: Philip Hofmann
|
| 50 |
+
License: CC-BY-4.0
|
| 51 |
+
Purpose: Pretrain
|
| 52 |
+
Subject: Realistic, Anime
|
| 53 |
+
Date: 10.04.2024
|
| 54 |
+
|
| 55 |
+
Dataset: [nomos_uni](https://github.com/muslll/neosr)
|
| 56 |
+
Dataset Size: 2989
|
| 57 |
+
OTF (on the fly augmentations): No
|
| 58 |
+
Pretrained Model: None
|
| 59 |
+
Iterations: 80'000
|
| 60 |
+
Batch Size: 12
|
| 61 |
+
GT Size: 128
|
| 62 |
+
|
| 63 |
+
Description: 4x SPAN pretrain trained on [neosr](https://github.com/muslll/neosr)'s [new adaptation of the multi-scale ssim loss](https://github.com/muslll/neosr/wiki/Losses#mssim_opt) from yesterdays update on downsampled nomos_uni dataset using kim's [dataset destroyer](https://github.com/Kim2091/helpful-scripts/tree/main/Dataset%20Destroyer) with down_up,linear,cubic_mitchell,lanczos,gauss,box (while down_up used the same and with range = 0.15,1.5).
|
| 64 |
+
The new augmentations except CutBlur have also been used (since CutBlur is meant to be applied to real-world SR and may cause undesired effects if applied to bicubic-only).
|
| 65 |
+
Config and training log provided for more details.
|
| 66 |
+
|
| 67 |
+
---
|
| 68 |
+
|
| 69 |
+
Showcase:
|
| 70 |
+
[7 Slowpics Examples](https://slow.pics/c/zyilXhKU)
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
|